Selection of roadmap for i386 platform End-of-Life (EOL)

[Stephen Welker]

On 17/04/2013, at 9:15 PM, John Marino <dragonflybsd at marino.st> wrote:

> The topic of how to cease i386 platform support has been discussed ad nauseam on the #dragonflybsd IRC channel.  I promised to post something on the mail lists as we got closer to 3.4 release.  I hope that we reach a conclusion rather than devolve into a never-ending and frustrating discussion.
> 
> For the purposes of discussion, assume that the EOL of the i386 platform will happen.  It's a question of "when" and "how", not "if".  Also, for the initial discussion's sake, let assume that EOL is defined as 31 December, 2014, approximately 19 months from now.
> 
> There are two schools of thought on the method of achieving the dropping of support for i386:
> 
> 1) Continue supporting i386 as usual for 3 more releases (e.g. 3.6, 3.8, 3.10) and then drop support completely (e.g. no new bug reports accepted).  One day it's fully supported, the next day it's not supported at all.
> 
> 2) Declare Release 3.4 as the last release for i386 but pledge to fix serious bugs and panics until the end of 2014.  Currently a release is only supported for about 6 months, so this would make Release 3.4 a kind of "Long-Term Support" release.  It would also receive periodic package updates until EOL.
> 
> My bias is towards approach #2.  From the perspective of a user, if their (older) box cpu is limited to the i386 architecture, then having extended support is probably more attractive than having the latest DragonFly technology.  From a developer point of view, it means a 50% decrease of support in some areas, including the architecture specific algorithms in the kernel.  Personally I'm also thinking about package building, non-base compiler bootstrapping, image building and mirroring, etc.  This can free up time to make the x86_64 platform better faster.
> 
> The main benefit to approach #1 is that Long Term Support can be avoided, which is primarily a benefit to (some) developers.  That is, it's easier to maintain a status quo than to fix bugs in a 1.5 year old release.  The benefit to users is that the last release of DragonFly for i386 would be more advanced than DragonFly 3.4, with the downside that it would also be unsupported (aka completely as-is, use at your own risk)
> 
> I believe that Release 3.4 will be a very good, stable release, and a worthy release to serve as a send-off for i386.  It's easily been the most stable version of DragonFly I've run, so I can imagine that serious bug-fix support won't be that taxing.
> 
> Anyway, the Project decisions I'd like to get out of this discussion with relatively little bloodshed is:
> 
> 1) Agreement on the EOL date (or Release if it's pegged to a release)
> 2) A declaration of which road map will be used (method #1 or #2)
> 
> I know some people might be tempted to argue to try to "save" the platform, but I think it's inevitable that it will be contracted. Again, I think it's merely a question of when and how base on these IRC discussions.

I would like to see DragonflyBSD 4.0 to be the beginning of x86_64 only releases. With the above mentioned restructure (or optimisation) of the kernel, many ABIs & APIs will change and 4.0 provides a milestone for that.

It could also mean that Hammer2 could be a headline feature of 4.0 - hopefully simplifying development.

People will also be aware that releases prior to 4.0 DragonflyBSD will support both x86_64 & i386.

Future releases of 3.x may add new features, but more importantly, provide some legacy users bug fixes and security updates.

regards,
Stephen Welker.